Publishing Surgeon Mortality at St. Alder
St. Alder Regional Hospital is considering publishing surgeon-level, risk-adjusted mortality rates for its five cardiac surgeons. One surgeon performs roughly three times as many high-complexity cases as colleagues and has a raw mortality rate of 4.1%, against a hospital average of 1.9%. After risk adjustment the gap narrows to 0.4 points and is not statistically significant at the hospital's volume. A neighbouring system that published comparable data saw a 14% decline in the high-risk cases it accepted over two years; some were transferred to academic centres, others were declined outright. Patient groups argue for publication. Several surgeons have said privately that they would stop accepting the most complex cases. The board has one meeting to decide, and state regulators have signalled they may mandate publication next year regardless.
1. Recommend what St. Alder should publish. Justify your recommendation with the statistical and ethical considerations in the setup.
Read the worked answer
Recommendation: publish, but not as a surgeon ranking. Publish risk-adjusted mortality by surgeon with confidence intervals, case-mix counts, and an explicit statement that the differences are not statistically distinguishable; publish process and team measures alongside. Refuse to issue a single-number league table.
The decisive fact is the 0.4-point gap that is not statistically significant at this volume. That sentence means the hospital cannot tell its five surgeons apart. A league table would present sampling noise as a quality rating, and patients would make irreversible decisions on it. Rate publication is only informative when the confidence intervals separate; here they do not.
The neighbouring system's 14% decline is the strongest evidence in the setup, and it cuts against naive publication. Some of those transfers were appropriate - high-risk cases belong at centres with more volume. But 'declined outright' is a harm, and 14% is large.
The ethical case still favours publishing something, because the alternative is patients deciding with no information and the state mandating a worse format in twelve months. The hospital's interest lies in pre-empting a crude mandate with a defensible version.
2. Patient groups say transparency is a right that outweighs statistical caution. How do you answer that argument?
Read the worked answer
The argument: patients are entitled to know their surgeon's outcomes; withholding data to protect surgeons from misinterpretation is paternalism dressed as statistics; and the 14% decline may simply be appropriate triage.
My answer, in two parts.
First, a right to information is a right to accurate information. A surgeon-level rate built on a handful of cases with non-overlapping uncertainty is not information about that surgeon; it is information about the hospital's sample size. Publishing it does not inform choice, it manufactures a false ranking. This is not a dodge: the same logic is why a lab reports a reference range rather than a single number with no error bar. Timely death and complication counts at the hospital level, waiting times, and volume by procedure would be genuinely informative, and I would publish those immediately.
Second, the autonomy argument actually requires more disclosure, not less. Patients deciding between a surgeon who takes on complex cases at 4.1% and one at 1.9% need to know that the first is operating on sicker patients. A raw rate with no case-mix context is more misleading than silence.
Where the patient groups are right: the board should not use 'patients will misread it' as a reason to publish nothing. That position is defensible only with a concrete alternative, which I have proposed.
3. What evidence would make surgeon-level publication defensible, and what would make you oppose publication entirely?
Read the worked answer
What would make it defensible: pooling three to five years of cases and applying a risk-adjustment model that has been validated against an external registry, to the point where confidence intervals around the five surgeons no longer overlap. If the intervals separate, the ranking is real and patients should see it. Volume is the binding constraint here - a single surgeon's annual case count is simply too small - so the defensible format is multi-year, and the hospital should say so.
What would make me oppose publication entirely: evidence that the adjustment model fails to capture the driver of complexity at this hospital. The 4.1% to 1.9% narrowing is only as good as the variables in the model. If the high-complexity surgeon takes cases defined by a factor the model omits - emergency status, reoperation, a specific anatomical category - then adjustment understates the gap and publication actively punishes the one surgeon willing to take those patients. Given that several surgeons have already said privately they would stop taking complex cases, a publication scheme that cannot be shown to distinguish skill from case selection will accelerate exactly the 14% effect without producing a single reliable ranking.
The test: if the model omits a documented complexity factor, I would publish hospital-level data only, and say why.